Mental note, I don’t think that adapt_model
automatically handles changing the number of channels based on the new data, just the number of outputs - we should look at modifying it to support that for audio, @zachcaceres @ste.
if batch norm causes problems, is there any point in using it with transfer learning?
If it’s in your pretrained model, then you don’t have any choice.
What is the best way to convert these models with custom heads into core ml model?
‘track_running_stats’ , saw this in pytorch … its helpful
track running stats is different: it’s to update the inner average mean/std. Jeremy was talking about training the learnable parameters.
oh ok , thanks …
You can export the PyTorch model and convert it to CoreML the same way as if it had not been customized.
Still trying to wrap my head around how I could use mixup on Tabular data…
taking backup of your good code helps
probably using embeddings, but not sure…
Along the same lines of what Jeremy is talking about right now - is it possible to have a TDD approach when doing deep learning?
Say more…
Linear combinations of numerics, datetimes, and one-hot encoded categorical variables seem pretty straightforward, no?
Is there any work around building open source simulators/debuggers that can help debug the kinds of issues that Jeremy just described without having to spend $$$?
I don’t even know if this is possible! Just curious.
Quote of the day (month? year?):
“Training models sucks. And deep learning is a miserable experience and you shouldn’t do it.” - Jeremy Howard
Reminder to upvote & ask questions here for the last 10 minutes of class:
For what it’s worth, I’ve heard Jeremy say this several times now: he’s worked a lot with TDD in the past, but in DL he seems to prefer working with notebooks which is in itself a kind of micro-TDD, if you have the discipline to check everything as you progress like he just mentioned.
What does Jeremy mean by scientific journal? Is it a file where all code goes by date? What is the best way to keep that?
What does Jeremy mean when he says “lab notes”? Is it a physical paper notebook he records stuff on? Or is there some software he’s using to record experiments/results/progress?
Edit: @maxim.pechyonkin same question at the same time!